-
Notifications
You must be signed in to change notification settings - Fork 14
Home
1flowbase is an open-source AI gateway for local agent clients. It lets Claude Code, Codex, OpenCode, Cline, Continue, and SDKs call one normal model endpoint while 1flowbase runs an observable workflow behind it.
Use 1flowbase when you want to publish a workflow as an OpenAI-compatible or Claude-compatible virtual model, combine multiple models behind one endpoint, mount vision or specialist models as tools, and inspect traces, tokens, latency, failures, and cost for every run.
Agent client -> one virtual model endpoint -> workflow -> trace / tokens / cost -> final answer
| Goal | Read this first |
|---|---|
| Install or upgrade 1flowbase | Repository quickstart |
| Make a text coding model understand screenshots | 让 GLM-5.2 在 Claude Code 里看图 / English |
| Publish a multi-model reviewer as one endpoint | Fusion 风格工作流 / English |
| Route one client endpoint across several models | 智能路由 / English |
| Understand token usage and agent cost | Claude Code Agent 的 Token BOM / English |
1flowbase is not only a model proxy and not only a workflow canvas.
| Category | Typical behavior | 1flowbase difference |
|---|---|---|
| LLM gateway / model router | Route one request to one provider or model | Compose multiple models, tools, verifiers, and formatters into one workflow-backed virtual model |
| AI workflow builder | Build an app-specific AI workflow | Publish that workflow as OpenAI-compatible and Claude-compatible model APIs |
| Agent framework | Code agent graphs in a framework | Provide a visual runtime, endpoint publishing, execution logs, and token visibility |
| Observability tool | Show traces or aggregate token totals | Connect cost and latency to workflow nodes, branch models, tool callbacks, and failures |
Use this when GLM-5.2, DeepSeek, or another strong text model should remain the main coding model, while images, screenshots, charts, PDFs, or UI mockups are routed to a mounted vision model.
Use this when a task benefits from independent model perspectives, branch review, synthesis, and visible token/cost tradeoffs.
Use this when clients should call one model name while 1flowbase handles routing, fallback, branch tools, provider choice, route reason, and trace logs.
Use this when Claude Code or another agent client feels expensive, slow, or opaque, and you need a per-run breakdown of model calls, cache-hit tokens, tool callbacks, routed work, latency, and failure state.